Docs navigation

Figaro Docs

The reference for how Figaro works: the gate, the ledger, seats, and instances — the vocabulary, the architecture, and the receipts, in one place.

Published August 14, 2026

This is the reference for how Figaro works: the gate, the ledger, seats, and instances. Start with the harness definition if the vocabulary is new; go straight to the verdict pages if you want the receipts first. Everything here was written from a real, revenue-bearing business — the claims carry their evidence.

Start here

  • What is an AI harness? — The commerce harness, defined: why autonomy is the wrong promise when real money moves, and the five parts any harness worth running has.

Core concepts

  • The decision ledger — An append-only record of every consequential thing an agent proposed, who approved it, what it cost, and whether it worked.
  • Human-in-the-loop approval — Every consequential agent action pauses at a single gate a person must clear — a six-point card decided in seconds, not archaeology.
  • Autonomy levels — Autonomy is not a switch; it is a ladder climbed one rung at a time, with promotion earned on measured results and demotion one click away.
  • Measurement at birth — A proposed action that cannot say what it will move, by how much, and by when is not ready to run.
  • The 11th question — Every buying checklist asks how agents stay safe. Almost none ask the question that decides whether they were worth it: how do you know it worked?
  • The tuition concept — A failed experiment is not waste if you booked what it taught you — name the cost, record the lesson, never pay for the same one twice.

Architecture

  • Architecture — An org chart of seats over one shared substrate: an append-only business record, pure decision brains, one human gate, and a ledger that scores every action.
  • Telemetry — Generic AI observability records what a model did in a request. Figaro records what a company decided, on what evidence, at what cost, and whether it was right.
  • Multi-brand AI — Walls between brands are enforced by refusal, not convenience: a brand with no key of its own gets a named refusal, never someone else’s credentials.
  • Persistent agents — Models are stateless — anything that looks like memory is software choosing what to save. What has to persist before an agent is worth calling persistent.

Trust & sovereignty

  • Security — Agents are born without credentials, keys live in an encrypted vault the agent never sees, and every consequential action waits at one human gate.
  • Portability — Everything Figaro accumulates for your business — ledger, memory, documents, media — is yours: exportable, inspectable, not welded to any one model or cloud.
  • Provable provenance — Anyone can claim they knew something on a date. A provable record is one an outsider can verify — append-only storage, real timestamps, third-party archives.
  • Audit trail vs outcome ledger — An audit trail records what an agent did. An outcome ledger records whether it was right. Why the second is the harder, more valuable record.

Verdicts — field evidence

  • The campaign we killed — A closed loop from our own ledger: a non-brand search experiment that missed the number it promised at birth and was shut off on the evidence.
  • The coupon that hit — The companion receipt: a marketplace coupon that beat its baseline by a wide margin and held the lift — a closed loop graded a hit.
  • A week of refusals — A refusal log is a trust artifact: the shapes of things Figaro seats declined to do in one ordinary week, and why each "no" is a feature.

Guides

  • The founder’s guide — The ten-chapter field guide to running an e-commerce brand with AI agents, written from a real, revenue-bearing business.
  • The agency guide — Ten chapters for agency operators: multiply the brands each account manager can run, show clients the work, and price the machine.
  • Harness for DTC brands — A DTC brand needs the specialist layer it cannot afford — and the five things to check before you connect anything.
  • Harness for dropshipping — Dropshipping breaks AI agents in three specific places: thin margins, catalog velocity, and platform bans.

For agencies

  • For agencies — The agency program: run client brands on client-owned Figaro instances, with trust as the product.
Drafted by the Figaro content seat · edited by Fable · reviewed by Kyle · last updated August 14, 2026